2 * Copyright (c) 1982, 1986, 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)sys_generic.c 8.5 (Berkeley) 1/21/94
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
40 #include "opt_compat.h"
41 #include "opt_ktrace.h"
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/sysproto.h>
46 #include <sys/filedesc.h>
47 #include <sys/filio.h>
48 #include <sys/fcntl.h>
51 #include <sys/signalvar.h>
52 #include <sys/socketvar.h>
54 #include <sys/kernel.h>
56 #include <sys/limits.h>
57 #include <sys/malloc.h>
59 #include <sys/resourcevar.h>
60 #include <sys/selinfo.h>
61 #include <sys/sleepqueue.h>
62 #include <sys/syscallsubr.h>
63 #include <sys/sysctl.h>
64 #include <sys/sysent.h>
65 #include <sys/vnode.h>
68 #include <sys/condvar.h>
70 #include <sys/ktrace.h>
73 #include <security/audit/audit.h>
75 static MALLOC_DEFINE(M_IOCTLOPS, "ioctlops", "ioctl data buffer");
76 static MALLOC_DEFINE(M_SELECT, "select", "select() buffer");
77 MALLOC_DEFINE(M_IOV, "iov", "large iov's");
79 static int pollscan(struct thread *, struct pollfd *, u_int);
80 static int pollrescan(struct thread *);
81 static int selscan(struct thread *, fd_mask **, fd_mask **, int);
82 static int selrescan(struct thread *, fd_mask **, fd_mask **);
83 static void selfdalloc(struct thread *, void *);
84 static void selfdfree(struct seltd *, struct selfd *);
85 static int dofileread(struct thread *, int, struct file *, struct uio *,
87 static int dofilewrite(struct thread *, int, struct file *, struct uio *,
89 static void doselwakeup(struct selinfo *, int);
90 static void seltdinit(struct thread *);
91 static int seltdwait(struct thread *, int);
92 static void seltdclear(struct thread *);
95 * One seltd per-thread allocated on demand as needed.
97 * t - protected by st_mtx
98 * k - Only accessed by curthread or read-only
101 STAILQ_HEAD(, selfd) st_selq; /* (k) List of selfds. */
102 struct selfd *st_free1; /* (k) free fd for read set. */
103 struct selfd *st_free2; /* (k) free fd for write set. */
104 struct mtx st_mtx; /* Protects struct seltd */
105 struct cv st_wait; /* (t) Wait channel. */
106 int st_flags; /* (t) SELTD_ flags. */
109 #define SELTD_PENDING 0x0001 /* We have pending events. */
110 #define SELTD_RESCAN 0x0002 /* Doing a rescan. */
113 * One selfd allocated per-thread per-file-descriptor.
114 * f - protected by sf_mtx
117 STAILQ_ENTRY(selfd) sf_link; /* (k) fds owned by this td. */
118 TAILQ_ENTRY(selfd) sf_threads; /* (f) fds on this selinfo. */
119 struct selinfo *sf_si; /* (f) selinfo when linked. */
120 struct mtx *sf_mtx; /* Pointer to selinfo mtx. */
121 struct seltd *sf_td; /* (k) owning seltd. */
122 void *sf_cookie; /* (k) fd or pollfd. */
125 static uma_zone_t selfd_zone;
127 #ifndef _SYS_SYSPROTO_H_
137 struct read_args *uap;
143 if (uap->nbyte > INT_MAX)
145 aiov.iov_base = uap->buf;
146 aiov.iov_len = uap->nbyte;
147 auio.uio_iov = &aiov;
149 auio.uio_resid = uap->nbyte;
150 auio.uio_segflg = UIO_USERSPACE;
151 error = kern_readv(td, uap->fd, &auio);
156 * Positioned read system call
158 #ifndef _SYS_SYSPROTO_H_
170 struct pread_args *uap;
176 if (uap->nbyte > INT_MAX)
178 aiov.iov_base = uap->buf;
179 aiov.iov_len = uap->nbyte;
180 auio.uio_iov = &aiov;
182 auio.uio_resid = uap->nbyte;
183 auio.uio_segflg = UIO_USERSPACE;
184 error = kern_preadv(td, uap->fd, &auio, uap->offset);
189 freebsd6_pread(td, uap)
191 struct freebsd6_pread_args *uap;
193 struct pread_args oargs;
196 oargs.buf = uap->buf;
197 oargs.nbyte = uap->nbyte;
198 oargs.offset = uap->offset;
199 return (pread(td, &oargs));
203 * Scatter read system call.
205 #ifndef _SYS_SYSPROTO_H_
213 readv(struct thread *td, struct readv_args *uap)
218 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
221 error = kern_readv(td, uap->fd, auio);
227 kern_readv(struct thread *td, int fd, struct uio *auio)
232 error = fget_read(td, fd, &fp);
235 error = dofileread(td, fd, fp, auio, (off_t)-1, 0);
241 * Scatter positioned read system call.
243 #ifndef _SYS_SYSPROTO_H_
252 preadv(struct thread *td, struct preadv_args *uap)
257 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
260 error = kern_preadv(td, uap->fd, auio, uap->offset);
266 kern_preadv(td, fd, auio, offset)
275 error = fget_read(td, fd, &fp);
278 if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
280 else if (offset < 0 && fp->f_vnode->v_type != VCHR)
283 error = dofileread(td, fd, fp, auio, offset, FOF_OFFSET);
289 * Common code for readv and preadv that reads data in
290 * from a file using the passed in uio, offset, and flags.
293 dofileread(td, fd, fp, auio, offset, flags)
304 struct uio *ktruio = NULL;
307 /* Finish zero length reads right here */
308 if (auio->uio_resid == 0) {
309 td->td_retval[0] = 0;
312 auio->uio_rw = UIO_READ;
313 auio->uio_offset = offset;
316 if (KTRPOINT(td, KTR_GENIO))
317 ktruio = cloneuio(auio);
319 cnt = auio->uio_resid;
320 if ((error = fo_read(fp, auio, td->td_ucred, flags, td))) {
321 if (auio->uio_resid != cnt && (error == ERESTART ||
322 error == EINTR || error == EWOULDBLOCK))
325 cnt -= auio->uio_resid;
327 if (ktruio != NULL) {
328 ktruio->uio_resid = cnt;
329 ktrgenio(fd, UIO_READ, ktruio, error);
332 td->td_retval[0] = cnt;
336 #ifndef _SYS_SYSPROTO_H_
346 struct write_args *uap;
352 if (uap->nbyte > INT_MAX)
354 aiov.iov_base = (void *)(uintptr_t)uap->buf;
355 aiov.iov_len = uap->nbyte;
356 auio.uio_iov = &aiov;
358 auio.uio_resid = uap->nbyte;
359 auio.uio_segflg = UIO_USERSPACE;
360 error = kern_writev(td, uap->fd, &auio);
365 * Positioned write system call.
367 #ifndef _SYS_SYSPROTO_H_
379 struct pwrite_args *uap;
385 if (uap->nbyte > INT_MAX)
387 aiov.iov_base = (void *)(uintptr_t)uap->buf;
388 aiov.iov_len = uap->nbyte;
389 auio.uio_iov = &aiov;
391 auio.uio_resid = uap->nbyte;
392 auio.uio_segflg = UIO_USERSPACE;
393 error = kern_pwritev(td, uap->fd, &auio, uap->offset);
398 freebsd6_pwrite(td, uap)
400 struct freebsd6_pwrite_args *uap;
402 struct pwrite_args oargs;
405 oargs.buf = uap->buf;
406 oargs.nbyte = uap->nbyte;
407 oargs.offset = uap->offset;
408 return (pwrite(td, &oargs));
412 * Gather write system call.
414 #ifndef _SYS_SYSPROTO_H_
422 writev(struct thread *td, struct writev_args *uap)
427 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
430 error = kern_writev(td, uap->fd, auio);
436 kern_writev(struct thread *td, int fd, struct uio *auio)
441 error = fget_write(td, fd, &fp);
444 error = dofilewrite(td, fd, fp, auio, (off_t)-1, 0);
450 * Gather positioned write system call.
452 #ifndef _SYS_SYSPROTO_H_
453 struct pwritev_args {
461 pwritev(struct thread *td, struct pwritev_args *uap)
466 error = copyinuio(uap->iovp, uap->iovcnt, &auio);
469 error = kern_pwritev(td, uap->fd, auio, uap->offset);
475 kern_pwritev(td, fd, auio, offset)
484 error = fget_write(td, fd, &fp);
487 if (!(fp->f_ops->fo_flags & DFLAG_SEEKABLE))
489 else if (offset < 0 && fp->f_vnode->v_type != VCHR)
492 error = dofilewrite(td, fd, fp, auio, offset, FOF_OFFSET);
498 * Common code for writev and pwritev that writes data to
499 * a file using the passed in uio, offset, and flags.
502 dofilewrite(td, fd, fp, auio, offset, flags)
513 struct uio *ktruio = NULL;
516 auio->uio_rw = UIO_WRITE;
518 auio->uio_offset = offset;
520 if (KTRPOINT(td, KTR_GENIO))
521 ktruio = cloneuio(auio);
523 cnt = auio->uio_resid;
524 if (fp->f_type == DTYPE_VNODE)
526 if ((error = fo_write(fp, auio, td->td_ucred, flags, td))) {
527 if (auio->uio_resid != cnt && (error == ERESTART ||
528 error == EINTR || error == EWOULDBLOCK))
530 /* Socket layer is responsible for issuing SIGPIPE. */
531 if (fp->f_type != DTYPE_SOCKET && error == EPIPE) {
532 PROC_LOCK(td->td_proc);
533 psignal(td->td_proc, SIGPIPE);
534 PROC_UNLOCK(td->td_proc);
537 cnt -= auio->uio_resid;
539 if (ktruio != NULL) {
540 ktruio->uio_resid = cnt;
541 ktrgenio(fd, UIO_WRITE, ktruio, error);
544 td->td_retval[0] = cnt;
549 * Truncate a file given a file descriptor.
551 * Can't use fget_write() here, since must return EINVAL and not EBADF if the
552 * descriptor isn't writable.
555 kern_ftruncate(td, fd, length)
566 error = fget(td, fd, &fp);
569 AUDIT_ARG(file, td->td_proc, fp);
570 if (!(fp->f_flag & FWRITE)) {
574 error = fo_truncate(fp, length, td->td_ucred, td);
579 #ifndef _SYS_SYSPROTO_H_
580 struct ftruncate_args {
589 struct ftruncate_args *uap;
592 return (kern_ftruncate(td, uap->fd, uap->length));
595 #if defined(COMPAT_43)
596 #ifndef _SYS_SYSPROTO_H_
597 struct oftruncate_args {
605 struct oftruncate_args *uap;
608 return (kern_ftruncate(td, uap->fd, uap->length));
610 #endif /* COMPAT_43 */
612 #ifndef _SYS_SYSPROTO_H_
621 ioctl(struct thread *td, struct ioctl_args *uap)
628 if (uap->com > 0xffffffff) {
630 "WARNING pid %d (%s): ioctl sign-extension ioctl %lx\n",
631 td->td_proc->p_pid, td->td_name, uap->com);
632 uap->com &= 0xffffffff;
637 * Interpret high order word to find amount of data to be
638 * copied to/from the user's address space.
640 size = IOCPARM_LEN(com);
641 if ((size > IOCPARM_MAX) ||
642 ((com & (IOC_VOID | IOC_IN | IOC_OUT)) == 0) ||
643 #if defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) || defined(COMPAT_43)
644 ((com & IOC_OUT) && size == 0) ||
646 ((com & (IOC_IN | IOC_OUT)) && size == 0) ||
648 ((com & IOC_VOID) && size > 0 && size != sizeof(int)))
652 if (!(com & IOC_VOID))
653 data = malloc((u_long)size, M_IOCTLOPS, M_WAITOK);
655 /* Integer argument. */
656 arg = (intptr_t)uap->data;
661 data = (void *)&uap->data;
663 error = copyin(uap->data, data, (u_int)size);
666 free(data, M_IOCTLOPS);
669 } else if (com & IOC_OUT) {
671 * Zero the buffer so the user always
672 * gets back something deterministic.
677 error = kern_ioctl(td, uap->fd, com, data);
679 if (error == 0 && (com & IOC_OUT))
680 error = copyout(data, uap->data, (u_int)size);
683 free(data, M_IOCTLOPS);
688 kern_ioctl(struct thread *td, int fd, u_long com, caddr_t data)
691 struct filedesc *fdp;
695 if ((error = fget(td, fd, &fp)) != 0)
697 if ((fp->f_flag & (FREAD | FWRITE)) == 0) {
701 fdp = td->td_proc->p_fd;
705 fdp->fd_ofileflags[fd] &= ~UF_EXCLOSE;
706 FILEDESC_XUNLOCK(fdp);
710 fdp->fd_ofileflags[fd] |= UF_EXCLOSE;
711 FILEDESC_XUNLOCK(fdp);
714 if ((tmp = *(int *)data))
715 atomic_set_int(&fp->f_flag, FNONBLOCK);
717 atomic_clear_int(&fp->f_flag, FNONBLOCK);
721 if ((tmp = *(int *)data))
722 atomic_set_int(&fp->f_flag, FASYNC);
724 atomic_clear_int(&fp->f_flag, FASYNC);
729 error = fo_ioctl(fp, com, data, td->td_ucred, td);
735 #ifndef _SYS_SYSPROTO_H_
738 fd_set *in, *ou, *ex;
744 register struct thread *td;
745 register struct select_args *uap;
747 struct timeval tv, *tvp;
750 if (uap->tv != NULL) {
751 error = copyin(uap->tv, &tv, sizeof(tv));
758 return (kern_select(td, uap->nd, uap->in, uap->ou, uap->ex, tvp));
762 kern_select(struct thread *td, int nd, fd_set *fd_in, fd_set *fd_ou,
763 fd_set *fd_ex, struct timeval *tvp)
765 struct filedesc *fdp;
767 * The magic 2048 here is chosen to be just enough for FD_SETSIZE
768 * infds with the new FD_SETSIZE of 1024, and more than enough for
769 * FD_SETSIZE infds, outfds and exceptfds with the old FD_SETSIZE
772 fd_mask s_selbits[howmany(2048, NFDBITS)];
773 fd_mask *ibits[3], *obits[3], *selbits, *sbp;
774 struct timeval atv, rtv, ttv;
776 u_int nbufbytes, ncpbytes, nfdbits;
780 fdp = td->td_proc->p_fd;
783 if (nd > td->td_proc->p_fd->fd_nfiles)
784 nd = td->td_proc->p_fd->fd_nfiles; /* forgiving; slightly wrong */
785 FILEDESC_SUNLOCK(fdp);
788 * Allocate just enough bits for the non-null fd_sets. Use the
789 * preallocated auto buffer if possible.
791 nfdbits = roundup(nd, NFDBITS);
792 ncpbytes = nfdbits / NBBY;
795 nbufbytes += 2 * ncpbytes;
797 nbufbytes += 2 * ncpbytes;
799 nbufbytes += 2 * ncpbytes;
800 if (nbufbytes <= sizeof s_selbits)
801 selbits = &s_selbits[0];
803 selbits = malloc(nbufbytes, M_SELECT, M_WAITOK);
806 * Assign pointers into the bit buffers and fetch the input bits.
807 * Put the output buffers together so that they can be bzeroed
811 #define getbits(name, x) \
816 ibits[x] = sbp + nbufbytes / 2 / sizeof *sbp; \
818 sbp += ncpbytes / sizeof *sbp; \
819 error = copyin(name, ibits[x], ncpbytes); \
829 bzero(selbits, nbufbytes / 2);
833 if (itimerfix(&atv)) {
837 getmicrouptime(&rtv);
838 timevaladd(&atv, &rtv);
845 /* Iterate until the timeout expires or descriptors become ready. */
847 error = selscan(td, ibits, obits, nd);
848 if (error || td->td_retval[0] != 0)
850 if (atv.tv_sec || atv.tv_usec) {
851 getmicrouptime(&rtv);
852 if (timevalcmp(&rtv, &atv, >=))
855 timevalsub(&ttv, &rtv);
856 timo = ttv.tv_sec > 24 * 60 * 60 ?
857 24 * 60 * 60 * hz : tvtohz(&ttv);
859 error = seltdwait(td, timo);
862 error = selrescan(td, ibits, obits);
863 if (error || td->td_retval[0] != 0)
869 /* select is not restarted after signals... */
870 if (error == ERESTART)
872 if (error == EWOULDBLOCK)
874 #define putbits(name, x) \
875 if (name && (error2 = copyout(obits[x], name, ncpbytes))) \
885 if (selbits != &s_selbits[0])
886 free(selbits, M_SELECT);
892 * Traverse the list of fds attached to this thread's seltd and check for
896 selrescan(struct thread *td, fd_mask **ibits, fd_mask **obits)
905 /* Note: backend also returns POLLHUP/POLLERR if appropriate. */
906 static int flag[3] = { POLLRDNORM, POLLWRNORM, POLLRDBAND };
907 struct filedesc *fdp = td->td_proc->p_fd;
911 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
912 fd = (int)(uintptr_t)sfp->sf_cookie;
915 /* If the selinfo wasn't cleared the event didn't fire. */
918 if ((fp = fget_locked(fdp, fd)) == NULL) {
919 FILEDESC_SUNLOCK(fdp);
922 for (msk = 0; msk < 3; msk++) {
923 if (ibits[msk] == NULL)
925 if ((ibits[msk][fd/NFDBITS] &
926 ((fd_mask) 1 << (fd % NFDBITS))) == 0)
928 if (fo_poll(fp, flag[msk], td->td_ucred, td)) {
929 obits[msk][(fd)/NFDBITS] |=
930 ((fd_mask)1 << ((fd) % NFDBITS));
935 FILEDESC_SUNLOCK(fdp);
937 td->td_retval[0] = n;
942 * Perform the initial filedescriptor scan and register ourselves with
946 selscan(td, ibits, obits, nfd)
948 fd_mask **ibits, **obits;
955 /* Note: backend also returns POLLHUP/POLLERR if appropriate. */
956 static int flag[3] = { POLLRDNORM, POLLWRNORM, POLLRDBAND };
957 struct filedesc *fdp = td->td_proc->p_fd;
960 for (msk = 0; msk < 3; msk++) {
961 if (ibits[msk] == NULL)
963 for (i = 0; i < nfd; i += NFDBITS) {
964 bits = ibits[msk][i/NFDBITS];
965 /* ffs(int mask) not portable, fd_mask is long */
966 for (fd = i; bits && fd < nfd; fd++, bits >>= 1) {
969 if ((fp = fget_locked(fdp, fd)) == NULL) {
970 FILEDESC_SUNLOCK(fdp);
973 selfdalloc(td, (void *)(uintptr_t)fd);
974 if (fo_poll(fp, flag[msk], td->td_ucred,
976 obits[msk][(fd)/NFDBITS] |=
977 ((fd_mask)1 << ((fd) % NFDBITS));
983 FILEDESC_SUNLOCK(fdp);
984 td->td_retval[0] = n;
988 #ifndef _SYS_SYSPROTO_H_
998 struct poll_args *uap;
1000 struct pollfd *bits;
1001 struct pollfd smallbits[32];
1002 struct timeval atv, rtv, ttv;
1003 int error = 0, timo;
1008 if (nfds > maxfilesperproc && nfds > FD_SETSIZE)
1010 ni = nfds * sizeof(struct pollfd);
1011 if (ni > sizeof(smallbits))
1012 bits = malloc(ni, M_TEMP, M_WAITOK);
1015 error = copyin(uap->fds, bits, ni);
1018 if (uap->timeout != INFTIM) {
1019 atv.tv_sec = uap->timeout / 1000;
1020 atv.tv_usec = (uap->timeout % 1000) * 1000;
1021 if (itimerfix(&atv)) {
1025 getmicrouptime(&rtv);
1026 timevaladd(&atv, &rtv);
1033 /* Iterate until the timeout expires or descriptors become ready. */
1035 error = pollscan(td, bits, nfds);
1036 if (error || td->td_retval[0] != 0)
1038 if (atv.tv_sec || atv.tv_usec) {
1039 getmicrouptime(&rtv);
1040 if (timevalcmp(&rtv, &atv, >=))
1043 timevalsub(&ttv, &rtv);
1044 timo = ttv.tv_sec > 24 * 60 * 60 ?
1045 24 * 60 * 60 * hz : tvtohz(&ttv);
1047 error = seltdwait(td, timo);
1050 error = pollrescan(td);
1051 if (error || td->td_retval[0] != 0)
1057 /* poll is not restarted after signals... */
1058 if (error == ERESTART)
1060 if (error == EWOULDBLOCK)
1063 error = copyout(bits, uap->fds, ni);
1068 if (ni > sizeof(smallbits))
1074 pollrescan(struct thread *td)
1080 struct filedesc *fdp;
1086 fdp = td->td_proc->p_fd;
1088 FILEDESC_SLOCK(fdp);
1089 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn) {
1090 fd = (struct pollfd *)sfp->sf_cookie;
1092 selfdfree(stp, sfp);
1093 /* If the selinfo wasn't cleared the event didn't fire. */
1096 fp = fdp->fd_ofiles[fd->fd];
1098 fd->revents = POLLNVAL;
1103 * Note: backend also returns POLLHUP and
1104 * POLLERR if appropriate.
1106 fd->revents = fo_poll(fp, fd->events, td->td_ucred, td);
1107 if (fd->revents != 0)
1110 FILEDESC_SUNLOCK(fdp);
1112 td->td_retval[0] = n;
1118 pollscan(td, fds, nfd)
1123 struct filedesc *fdp = td->td_proc->p_fd;
1128 FILEDESC_SLOCK(fdp);
1129 for (i = 0; i < nfd; i++, fds++) {
1130 if (fds->fd >= fdp->fd_nfiles) {
1131 fds->revents = POLLNVAL;
1133 } else if (fds->fd < 0) {
1136 fp = fdp->fd_ofiles[fds->fd];
1138 fds->revents = POLLNVAL;
1142 * Note: backend also returns POLLHUP and
1143 * POLLERR if appropriate.
1145 selfdalloc(td, fds);
1146 fds->revents = fo_poll(fp, fds->events,
1148 if (fds->revents != 0)
1153 FILEDESC_SUNLOCK(fdp);
1154 td->td_retval[0] = n;
1159 * OpenBSD poll system call.
1161 * XXX this isn't quite a true representation.. OpenBSD uses select ops.
1163 #ifndef _SYS_SYSPROTO_H_
1164 struct openbsd_poll_args {
1171 openbsd_poll(td, uap)
1172 register struct thread *td;
1173 register struct openbsd_poll_args *uap;
1175 return (poll(td, (struct poll_args *)uap));
1179 * XXX This was created specifically to support netncp and netsmb. This
1180 * allows the caller to specify a socket to wait for events on. It returns
1181 * 0 if any events matched and an error otherwise. There is no way to
1182 * determine which events fired.
1185 selsocket(struct socket *so, int events, struct timeval *tvp, struct thread *td)
1187 struct timeval atv, rtv, ttv;
1192 if (itimerfix(&atv))
1194 getmicrouptime(&rtv);
1195 timevaladd(&atv, &rtv);
1204 * Iterate until the timeout expires or the socket becomes ready.
1207 selfdalloc(td, NULL);
1208 error = sopoll(so, events, NULL, td);
1209 /* error here is actually the ready events. */
1212 if (atv.tv_sec || atv.tv_usec) {
1213 getmicrouptime(&rtv);
1214 if (timevalcmp(&rtv, &atv, >=)) {
1216 return (EWOULDBLOCK);
1219 timevalsub(&ttv, &rtv);
1220 timo = ttv.tv_sec > 24 * 60 * 60 ?
1221 24 * 60 * 60 * hz : tvtohz(&ttv);
1223 error = seltdwait(td, timo);
1228 /* XXX Duplicates ncp/smb behavior. */
1229 if (error == ERESTART)
1235 * Preallocate two selfds associated with 'cookie'. Some fo_poll routines
1236 * have two select sets, one for read and another for write.
1239 selfdalloc(struct thread *td, void *cookie)
1244 if (stp->st_free1 == NULL)
1245 stp->st_free1 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
1246 stp->st_free1->sf_td = stp;
1247 stp->st_free1->sf_cookie = cookie;
1248 if (stp->st_free2 == NULL)
1249 stp->st_free2 = uma_zalloc(selfd_zone, M_WAITOK|M_ZERO);
1250 stp->st_free2->sf_td = stp;
1251 stp->st_free2->sf_cookie = cookie;
1255 selfdfree(struct seltd *stp, struct selfd *sfp)
1257 STAILQ_REMOVE(&stp->st_selq, sfp, selfd, sf_link);
1258 mtx_lock(sfp->sf_mtx);
1260 TAILQ_REMOVE(&sfp->sf_si->si_tdlist, sfp, sf_threads);
1261 mtx_unlock(sfp->sf_mtx);
1262 uma_zfree(selfd_zone, sfp);
1266 * Record a select request.
1269 selrecord(selector, sip)
1270 struct thread *selector;
1271 struct selinfo *sip;
1277 stp = selector->td_sel;
1279 * Don't record when doing a rescan.
1281 if (stp->st_flags & SELTD_RESCAN)
1284 * Grab one of the preallocated descriptors.
1287 if ((sfp = stp->st_free1) != NULL)
1288 stp->st_free1 = NULL;
1289 else if ((sfp = stp->st_free2) != NULL)
1290 stp->st_free2 = NULL;
1292 panic("selrecord: No free selfd on selq");
1293 mtxp = mtx_pool_find(mtxpool_sleep, sip);
1295 * Initialize the sfp and queue it in the thread.
1299 STAILQ_INSERT_TAIL(&stp->st_selq, sfp, sf_link);
1301 * Now that we've locked the sip, check for initialization.
1304 if (sip->si_mtx == NULL) {
1306 TAILQ_INIT(&sip->si_tdlist);
1309 * Add this thread to the list of selfds listening on this selinfo.
1311 TAILQ_INSERT_TAIL(&sip->si_tdlist, sfp, sf_threads);
1312 mtx_unlock(sip->si_mtx);
1315 /* Wake up a selecting thread. */
1318 struct selinfo *sip;
1320 doselwakeup(sip, -1);
1323 /* Wake up a selecting thread, and set its priority. */
1325 selwakeuppri(sip, pri)
1326 struct selinfo *sip;
1329 doselwakeup(sip, pri);
1333 * Do a wakeup when a selectable event occurs.
1336 doselwakeup(sip, pri)
1337 struct selinfo *sip;
1344 /* If it's not initialized there can't be any waiters. */
1345 if (sip->si_mtx == NULL)
1348 * Locking the selinfo locks all selfds associated with it.
1350 mtx_lock(sip->si_mtx);
1351 TAILQ_FOREACH_SAFE(sfp, &sip->si_tdlist, sf_threads, sfn) {
1353 * Once we remove this sfp from the list and clear the
1354 * sf_si seltdclear will know to ignore this si.
1356 TAILQ_REMOVE(&sip->si_tdlist, sfp, sf_threads);
1359 mtx_lock(&stp->st_mtx);
1360 stp->st_flags |= SELTD_PENDING;
1361 cv_broadcastpri(&stp->st_wait, pri);
1362 mtx_unlock(&stp->st_mtx);
1364 mtx_unlock(sip->si_mtx);
1368 seltdinit(struct thread *td)
1372 if ((stp = td->td_sel) != NULL)
1374 td->td_sel = stp = malloc(sizeof(*stp), M_SELECT, M_WAITOK|M_ZERO);
1375 mtx_init(&stp->st_mtx, "sellck", NULL, MTX_DEF);
1376 cv_init(&stp->st_wait, "select");
1379 STAILQ_INIT(&stp->st_selq);
1383 seltdwait(struct thread *td, int timo)
1390 * An event of interest may occur while we do not hold the seltd
1391 * locked so check the pending flag before we sleep.
1393 mtx_lock(&stp->st_mtx);
1395 * Any further calls to selrecord will be a rescan.
1397 stp->st_flags |= SELTD_RESCAN;
1398 if (stp->st_flags & SELTD_PENDING) {
1399 mtx_unlock(&stp->st_mtx);
1403 error = cv_timedwait_sig(&stp->st_wait, &stp->st_mtx, timo);
1405 error = cv_wait_sig(&stp->st_wait, &stp->st_mtx);
1406 mtx_unlock(&stp->st_mtx);
1412 seltdfini(struct thread *td)
1420 uma_zfree(selfd_zone, stp->st_free1);
1422 uma_zfree(selfd_zone, stp->st_free2);
1424 free(stp, M_SELECT);
1428 * Remove the references to the thread from all of the objects we were
1432 seltdclear(struct thread *td)
1439 STAILQ_FOREACH_SAFE(sfp, &stp->st_selq, sf_link, sfn)
1440 selfdfree(stp, sfp);
1444 static void selectinit(void *);
1445 SYSINIT(select, SI_SUB_SYSCALLS, SI_ORDER_ANY, selectinit, NULL);
1447 selectinit(void *dummy __unused)
1449 selfd_zone = uma_zcreate("selfd", sizeof(struct selfd), NULL, NULL,
1450 NULL, NULL, UMA_ALIGN_PTR, 0);